Modeling layout design for multiple-view visualization via Bayesian inference

Layout design for multiple-view visualization (MV) concerns primarily how to arrange views in layouts that are geometrically and topologically plausible. Guidelines for MV layout design suggest considerations on various design factors, including view (e.g., bar and line charts), viewport (e.g., mobile vs. desktop), and coordination (e.g., exploration vs. comparison), along with expertise and preference of the designer. Recent studies have revealed the diverse space of MV layout design via statistical analysis on empirical MVs, yet neglect the effects of those design factors. To address the gap, this work proposes to model the effects of design factors on MV layouts via Bayesian probabilistic inference. Specifically, we access three important properties of MV layout, i.e., maximum area ratio and weighted average aspect ratio as geometric metrics, and layout topology as a topological metric. We update the posterior probability of layout metrics given design factors by penetrating MVs from recent visualization publications. The analyses reveal many insightful MV layout design patterns, such as views in coordination type of comparison exhibit more balanced area ratio, while those for exploration are more scattered. This work makes a prominent starting point for a thorough understanding of MV layout design patterns. On the basis, we discuss how practitioners can use Bayesian inference approach for future research on finer-annotated visualization datasets and more comprehensive design factors and properties.